hm-rnn
Implementation of a hierarchical multiscale recurrent neural network, following (Chung, Ahn & Bengio, 2016, https://arxiv.org/abs/1609.01704).
This implementation uses plain RNNs (in contrast to the LSTMs in the original paper). It uses Python 3.6 and was tested with PyTorch 0.41 and scikit-learn 0.19.1.
If you want to learn how to create a new, custom environment to install the required versions of Python, PyTorch and scikit-learn using conda, please look at: https://conda.io/docs/user-guide/tasks/manage-environments.html For a getting started guide to PyTorch look here: https://pytorch.org/get-started/locally/
To start training the model, get some training dataset (in the form of a single .txt file) and put the path in the corresponding line of train.py. Then you can start training by executing train.py. It will print a segmentation of some training data and some samples from the learned model every 100th training step.